Multilingual intent matching engine

1. A method, comprising:

accessing a natural language query corresponding to one of a plurality of natural languages;

facilitating a mapping of the natural language query to a vector using a query-to-vector engine configured to leverage word embeddings in each of the plurality of natural languages to map natural language queries in the plurality of natural languages to vectors corresponding to meanings of the natural language queries;

matching, using a vector-to-intent engine, the vector to an intent representing a prediction associated with the natural language query, the vector-to-intent engine being trained by locking the word embeddings and using supervised learning with a dataset comprising a set of queries and intents matched to queries in the set; and

providing a response to the natural language query based on the intent.

2. The method of claim 1 , comprising:

testing the query-to-vector engine and a vector-to-intent engine used to match the vector to the intent by verifying that a first query in a first natural language and a translation of the first query into a second natural language match to a same intent.

3. The method of claim 1 , wherein the intent represents a grouping of a set of queries, including the natural language query, for further processing.

4. The method of claim 1 , wherein the vector is a numeric vector in a multi-dimensional space.

5. The method of claim 1 , wherein facilitating the mapping of the natural language query to the vector does not include translating the natural language query into a natural language different from a natural language of the natural language query.

6. The method of claim 1 , wherein the vector is matched to the intent using a machine learning technique.

7. The method of claim 1 , wherein the vector is matched to the intent using a statistical technique.

8. The method of claim 1 , wherein locking the word embeddings comprises forgoing updating the word embeddings via stochastic gradient descent, wherein the word embeddings are updated during operation of the query-to-vector engine when the word embeddings are not locked.

9. The method of claim 1 , wherein locking the word embeddings comprises forgoing updating the word embeddings via online learning, wherein the word embeddings are updated during operation of the query-to-vector engine when the word embeddings are not locked.

10. A non-transitory computer readable medium storing instructions operable to cause one or more processors to perform operations comprising:

accessing a natural language query corresponding to one of a plurality of natural languages;

facilitating a mapping of the natural language query to a vector using a query-to-vector engine configured to leverage word embeddings in each of the plurality of natural languages to map natural language queries in the plurality of natural languages to vectors corresponding to meanings of the natural language queries;

matching, using a vector-to-intent engine, the vector to an intent representing a prediction associated with the natural language query, the vector-to-intent engine being trained by locking the word embeddings and using supervised learning with a dataset comprising a set of queries and intents matched to queries in the set; and

providing a response to the natural language query based on the intent.

11. The computer readable medium of claim 10 , the operations comprising:

testing the query-to-vector engine and a machine learning engine used to match the vector to the intent by verifying that a first query in a first natural language and a translation of the first query into a second natural language match to a same intent.

12. The computer readable medium of claim 10 , wherein the intent represents a grouping of a set of queries, including the natural language query, for further processing using a specified technique from a set of techniques.

13. The computer readable medium of claim 10 , wherein the vector represents a coordinate within a multi-dimensional space.

14. The computer readable medium of claim 10 , wherein facilitating the mapping of the natural language query to the vector does not include machine translation of the natural language query.

15. An apparatus comprising:

a memory; and

a processor configured to execute instructions stored in the memory to:

access a natural language query corresponding to one of a plurality of natural languages;

facilitate a mapping of the natural language query to a vector using a query-to-vector engine configured to leverage word embeddings in each of the plurality of natural languages to map natural language queries in the plurality of natural languages to vectors corresponding to meanings of the natural language queries;

match, using a vector-to-intent engine, the vector to an intent representing a prediction associated with the natural language query, the vector-to-intent engine being trained by locking the word embeddings and using supervised learning with a dataset comprising a set of queries and intents matched to queries in the set; and

provide a response to the natural language query based on the intent.

16. The apparatus of claim 15 , the processor configured to execute the instructions stored in the memory to:

test the query-to-vector engine and an artificial intelligence engine used to match the vector to the intent by verifying that a first query in a first natural language and a translation of the first query into a second natural language match to a same intent.

17. The apparatus of claim 15 , wherein the intent represents a grouping of a set of queries, including the natural language query, for further computer-implemented processing using a specified technique.

18. The apparatus of claim 15 , wherein facilitating the mapping of the natural language query to the vector does not include machine translation.

19. The apparatus of claim 15 , wherein the vector is matched to the intent using an artificial intelligence technique.